Root/fs/ocfs2/dlmfs/dlmfs.c

1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmfs.c
5 *
6 * Code which implements the kernel side of a minimal userspace
7 * interface to our DLM. This file handles the virtual file system
8 * used for communication with userspace. Credit should go to ramfs,
9 * which was a template for the fs side of this module.
10 *
11 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public
15 * License as published by the Free Software Foundation; either
16 * version 2 of the License, or (at your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * General Public License for more details.
22 *
23 * You should have received a copy of the GNU General Public
24 * License along with this program; if not, write to the
25 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
26 * Boston, MA 021110-1307, USA.
27 */
28
29/* Simple VFS hooks based on: */
30/*
31 * Resizable simple ram filesystem for Linux.
32 *
33 * Copyright (C) 2000 Linus Torvalds.
34 * 2000 Transmeta Corp.
35 */
36
37#include <linux/module.h>
38#include <linux/fs.h>
39#include <linux/pagemap.h>
40#include <linux/types.h>
41#include <linux/slab.h>
42#include <linux/highmem.h>
43#include <linux/init.h>
44#include <linux/string.h>
45#include <linux/backing-dev.h>
46#include <linux/poll.h>
47
48#include <asm/uaccess.h>
49
50#include "stackglue.h"
51#include "userdlm.h"
52#include "dlmfsver.h"
53
54#define MLOG_MASK_PREFIX ML_DLMFS
55#include "cluster/masklog.h"
56
57
58static const struct super_operations dlmfs_ops;
59static const struct file_operations dlmfs_file_operations;
60static const struct inode_operations dlmfs_dir_inode_operations;
61static const struct inode_operations dlmfs_root_inode_operations;
62static const struct inode_operations dlmfs_file_inode_operations;
63static struct kmem_cache *dlmfs_inode_cache;
64
65struct workqueue_struct *user_dlm_worker;
66
67
68
69/*
70 * These are the ABI capabilities of dlmfs.
71 *
72 * Over time, dlmfs has added some features that were not part of the
73 * initial ABI. Unfortunately, some of these features are not detectable
74 * via standard usage. For example, Linux's default poll always returns
75 * POLLIN, so there is no way for a caller of poll(2) to know when dlmfs
76 * added poll support. Instead, we provide this list of new capabilities.
77 *
78 * Capabilities is a read-only attribute. We do it as a module parameter
79 * so we can discover it whether dlmfs is built in, loaded, or even not
80 * loaded.
81 *
82 * The ABI features are local to this machine's dlmfs mount. This is
83 * distinct from the locking protocol, which is concerned with inter-node
84 * interaction.
85 *
86 * Capabilities:
87 * - bast : POLLIN against the file descriptor of a held lock
88 * signifies a bast fired on the lock.
89 */
90#define DLMFS_CAPABILITIES "bast stackglue"
91extern int param_set_dlmfs_capabilities(const char *val,
92                    struct kernel_param *kp)
93{
94    printk(KERN_ERR "%s: readonly parameter\n", kp->name);
95    return -EINVAL;
96}
97static int param_get_dlmfs_capabilities(char *buffer,
98                    struct kernel_param *kp)
99{
100    return strlcpy(buffer, DLMFS_CAPABILITIES,
101               strlen(DLMFS_CAPABILITIES) + 1);
102}
103module_param_call(capabilities, param_set_dlmfs_capabilities,
104          param_get_dlmfs_capabilities, NULL, 0444);
105MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
106
107
108/*
109 * decodes a set of open flags into a valid lock level and a set of flags.
110 * returns < 0 if we have invalid flags
111 * flags which mean something to us:
112 * O_RDONLY -> PRMODE level
113 * O_WRONLY -> EXMODE level
114 *
115 * O_NONBLOCK -> LKM_NOQUEUE
116 */
117static int dlmfs_decode_open_flags(int open_flags,
118                   int *level,
119                   int *flags)
120{
121    if (open_flags & (O_WRONLY|O_RDWR))
122        *level = LKM_EXMODE;
123    else
124        *level = LKM_PRMODE;
125
126    *flags = 0;
127    if (open_flags & O_NONBLOCK)
128        *flags |= LKM_NOQUEUE;
129
130    return 0;
131}
132
133static int dlmfs_file_open(struct inode *inode,
134               struct file *file)
135{
136    int status, level, flags;
137    struct dlmfs_filp_private *fp = NULL;
138    struct dlmfs_inode_private *ip;
139
140    if (S_ISDIR(inode->i_mode))
141        BUG();
142
143    mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
144        file->f_flags);
145
146    status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
147    if (status < 0)
148        goto bail;
149
150    /* We don't want to honor O_APPEND at read/write time as it
151     * doesn't make sense for LVB writes. */
152    file->f_flags &= ~O_APPEND;
153
154    fp = kmalloc(sizeof(*fp), GFP_NOFS);
155    if (!fp) {
156        status = -ENOMEM;
157        goto bail;
158    }
159    fp->fp_lock_level = level;
160
161    ip = DLMFS_I(inode);
162
163    status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
164    if (status < 0) {
165        /* this is a strange error to return here but I want
166         * to be able userspace to be able to distinguish a
167         * valid lock request from one that simply couldn't be
168         * granted. */
169        if (flags & LKM_NOQUEUE && status == -EAGAIN)
170            status = -ETXTBSY;
171        kfree(fp);
172        goto bail;
173    }
174
175    file->private_data = fp;
176bail:
177    return status;
178}
179
180static int dlmfs_file_release(struct inode *inode,
181                  struct file *file)
182{
183    int level, status;
184    struct dlmfs_inode_private *ip = DLMFS_I(inode);
185    struct dlmfs_filp_private *fp =
186        (struct dlmfs_filp_private *) file->private_data;
187
188    if (S_ISDIR(inode->i_mode))
189        BUG();
190
191    mlog(0, "close called on inode %lu\n", inode->i_ino);
192
193    status = 0;
194    if (fp) {
195        level = fp->fp_lock_level;
196        if (level != LKM_IVMODE)
197            user_dlm_cluster_unlock(&ip->ip_lockres, level);
198
199        kfree(fp);
200        file->private_data = NULL;
201    }
202
203    return 0;
204}
205
206/*
207 * We do ->setattr() just to override size changes. Our size is the size
208 * of the LVB and nothing else.
209 */
210static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
211{
212    int error;
213    struct inode *inode = dentry->d_inode;
214
215    attr->ia_valid &= ~ATTR_SIZE;
216    error = inode_change_ok(inode, attr);
217    if (!error)
218        error = inode_setattr(inode, attr);
219
220    return error;
221}
222
223static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait)
224{
225    int event = 0;
226    struct inode *inode = file->f_path.dentry->d_inode;
227    struct dlmfs_inode_private *ip = DLMFS_I(inode);
228
229    poll_wait(file, &ip->ip_lockres.l_event, wait);
230
231    spin_lock(&ip->ip_lockres.l_lock);
232    if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
233        event = POLLIN | POLLRDNORM;
234    spin_unlock(&ip->ip_lockres.l_lock);
235
236    return event;
237}
238
239static ssize_t dlmfs_file_read(struct file *filp,
240                   char __user *buf,
241                   size_t count,
242                   loff_t *ppos)
243{
244    int bytes_left;
245    ssize_t readlen, got;
246    char *lvb_buf;
247    struct inode *inode = filp->f_path.dentry->d_inode;
248
249    mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
250        inode->i_ino, count, *ppos);
251
252    if (*ppos >= i_size_read(inode))
253        return 0;
254
255    if (!count)
256        return 0;
257
258    if (!access_ok(VERIFY_WRITE, buf, count))
259        return -EFAULT;
260
261    /* don't read past the lvb */
262    if ((count + *ppos) > i_size_read(inode))
263        readlen = i_size_read(inode) - *ppos;
264    else
265        readlen = count - *ppos;
266
267    lvb_buf = kmalloc(readlen, GFP_NOFS);
268    if (!lvb_buf)
269        return -ENOMEM;
270
271    got = user_dlm_read_lvb(inode, lvb_buf, readlen);
272    if (got) {
273        BUG_ON(got != readlen);
274        bytes_left = __copy_to_user(buf, lvb_buf, readlen);
275        readlen -= bytes_left;
276    } else
277        readlen = 0;
278
279    kfree(lvb_buf);
280
281    *ppos = *ppos + readlen;
282
283    mlog(0, "read %zd bytes\n", readlen);
284    return readlen;
285}
286
287static ssize_t dlmfs_file_write(struct file *filp,
288                const char __user *buf,
289                size_t count,
290                loff_t *ppos)
291{
292    int bytes_left;
293    ssize_t writelen;
294    char *lvb_buf;
295    struct inode *inode = filp->f_path.dentry->d_inode;
296
297    mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
298        inode->i_ino, count, *ppos);
299
300    if (*ppos >= i_size_read(inode))
301        return -ENOSPC;
302
303    if (!count)
304        return 0;
305
306    if (!access_ok(VERIFY_READ, buf, count))
307        return -EFAULT;
308
309    /* don't write past the lvb */
310    if ((count + *ppos) > i_size_read(inode))
311        writelen = i_size_read(inode) - *ppos;
312    else
313        writelen = count - *ppos;
314
315    lvb_buf = kmalloc(writelen, GFP_NOFS);
316    if (!lvb_buf)
317        return -ENOMEM;
318
319    bytes_left = copy_from_user(lvb_buf, buf, writelen);
320    writelen -= bytes_left;
321    if (writelen)
322        user_dlm_write_lvb(inode, lvb_buf, writelen);
323
324    kfree(lvb_buf);
325
326    *ppos = *ppos + writelen;
327    mlog(0, "wrote %zd bytes\n", writelen);
328    return writelen;
329}
330
331static void dlmfs_init_once(void *foo)
332{
333    struct dlmfs_inode_private *ip =
334        (struct dlmfs_inode_private *) foo;
335
336    ip->ip_conn = NULL;
337    ip->ip_parent = NULL;
338
339    inode_init_once(&ip->ip_vfs_inode);
340}
341
342static struct inode *dlmfs_alloc_inode(struct super_block *sb)
343{
344    struct dlmfs_inode_private *ip;
345
346    ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
347    if (!ip)
348        return NULL;
349
350    return &ip->ip_vfs_inode;
351}
352
353static void dlmfs_destroy_inode(struct inode *inode)
354{
355    kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
356}
357
358static void dlmfs_clear_inode(struct inode *inode)
359{
360    int status;
361    struct dlmfs_inode_private *ip;
362
363    if (!inode)
364        return;
365
366    mlog(0, "inode %lu\n", inode->i_ino);
367
368    ip = DLMFS_I(inode);
369
370    if (S_ISREG(inode->i_mode)) {
371        status = user_dlm_destroy_lock(&ip->ip_lockres);
372        if (status < 0)
373            mlog_errno(status);
374        iput(ip->ip_parent);
375        goto clear_fields;
376    }
377
378    mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
379    /* we must be a directory. If required, lets unregister the
380     * dlm context now. */
381    if (ip->ip_conn)
382        user_dlm_unregister(ip->ip_conn);
383clear_fields:
384    ip->ip_parent = NULL;
385    ip->ip_conn = NULL;
386}
387
388static struct backing_dev_info dlmfs_backing_dev_info = {
389    .name = "ocfs2-dlmfs",
390    .ra_pages = 0, /* No readahead */
391    .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
392};
393
394static struct inode *dlmfs_get_root_inode(struct super_block *sb)
395{
396    struct inode *inode = new_inode(sb);
397    int mode = S_IFDIR | 0755;
398    struct dlmfs_inode_private *ip;
399
400    if (inode) {
401        ip = DLMFS_I(inode);
402
403        inode->i_mode = mode;
404        inode->i_uid = current_fsuid();
405        inode->i_gid = current_fsgid();
406        inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
407        inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
408        inc_nlink(inode);
409
410        inode->i_fop = &simple_dir_operations;
411        inode->i_op = &dlmfs_root_inode_operations;
412    }
413
414    return inode;
415}
416
417static struct inode *dlmfs_get_inode(struct inode *parent,
418                     struct dentry *dentry,
419                     int mode)
420{
421    struct super_block *sb = parent->i_sb;
422    struct inode * inode = new_inode(sb);
423    struct dlmfs_inode_private *ip;
424
425    if (!inode)
426        return NULL;
427
428    inode->i_mode = mode;
429    inode->i_uid = current_fsuid();
430    inode->i_gid = current_fsgid();
431    inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
432    inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
433
434    ip = DLMFS_I(inode);
435    ip->ip_conn = DLMFS_I(parent)->ip_conn;
436
437    switch (mode & S_IFMT) {
438    default:
439        /* for now we don't support anything other than
440         * directories and regular files. */
441        BUG();
442        break;
443    case S_IFREG:
444        inode->i_op = &dlmfs_file_inode_operations;
445        inode->i_fop = &dlmfs_file_operations;
446
447        i_size_write(inode, DLM_LVB_LEN);
448
449        user_dlm_lock_res_init(&ip->ip_lockres, dentry);
450
451        /* released at clear_inode time, this insures that we
452         * get to drop the dlm reference on each lock *before*
453         * we call the unregister code for releasing parent
454         * directories. */
455        ip->ip_parent = igrab(parent);
456        BUG_ON(!ip->ip_parent);
457        break;
458    case S_IFDIR:
459        inode->i_op = &dlmfs_dir_inode_operations;
460        inode->i_fop = &simple_dir_operations;
461
462        /* directory inodes start off with i_nlink ==
463         * 2 (for "." entry) */
464        inc_nlink(inode);
465        break;
466    }
467
468    if (parent->i_mode & S_ISGID) {
469        inode->i_gid = parent->i_gid;
470        if (S_ISDIR(mode))
471            inode->i_mode |= S_ISGID;
472    }
473
474    return inode;
475}
476
477/*
478 * File creation. Allocate an inode, and we're done..
479 */
480/* SMP-safe */
481static int dlmfs_mkdir(struct inode * dir,
482               struct dentry * dentry,
483               int mode)
484{
485    int status;
486    struct inode *inode = NULL;
487    struct qstr *domain = &dentry->d_name;
488    struct dlmfs_inode_private *ip;
489    struct ocfs2_cluster_connection *conn;
490
491    mlog(0, "mkdir %.*s\n", domain->len, domain->name);
492
493    /* verify that we have a proper domain */
494    if (domain->len >= GROUP_NAME_MAX) {
495        status = -EINVAL;
496        mlog(ML_ERROR, "invalid domain name for directory.\n");
497        goto bail;
498    }
499
500    inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
501    if (!inode) {
502        status = -ENOMEM;
503        mlog_errno(status);
504        goto bail;
505    }
506
507    ip = DLMFS_I(inode);
508
509    conn = user_dlm_register(domain);
510    if (IS_ERR(conn)) {
511        status = PTR_ERR(conn);
512        mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
513             status, domain->len, domain->name);
514        goto bail;
515    }
516    ip->ip_conn = conn;
517
518    inc_nlink(dir);
519    d_instantiate(dentry, inode);
520    dget(dentry); /* Extra count - pin the dentry in core */
521
522    status = 0;
523bail:
524    if (status < 0)
525        iput(inode);
526    return status;
527}
528
529static int dlmfs_create(struct inode *dir,
530            struct dentry *dentry,
531            int mode,
532            struct nameidata *nd)
533{
534    int status = 0;
535    struct inode *inode;
536    struct qstr *name = &dentry->d_name;
537
538    mlog(0, "create %.*s\n", name->len, name->name);
539
540    /* verify name is valid and doesn't contain any dlm reserved
541     * characters */
542    if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
543        name->name[0] == '$') {
544        status = -EINVAL;
545        mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
546             name->name);
547        goto bail;
548    }
549
550    inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
551    if (!inode) {
552        status = -ENOMEM;
553        mlog_errno(status);
554        goto bail;
555    }
556
557    d_instantiate(dentry, inode);
558    dget(dentry); /* Extra count - pin the dentry in core */
559bail:
560    return status;
561}
562
563static int dlmfs_unlink(struct inode *dir,
564            struct dentry *dentry)
565{
566    int status;
567    struct inode *inode = dentry->d_inode;
568
569    mlog(0, "unlink inode %lu\n", inode->i_ino);
570
571    /* if there are no current holders, or none that are waiting
572     * to acquire a lock, this basically destroys our lockres. */
573    status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
574    if (status < 0) {
575        mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
576             dentry->d_name.len, dentry->d_name.name, status);
577        goto bail;
578    }
579    status = simple_unlink(dir, dentry);
580bail:
581    return status;
582}
583
584static int dlmfs_fill_super(struct super_block * sb,
585                void * data,
586                int silent)
587{
588    struct inode * inode;
589    struct dentry * root;
590
591    sb->s_maxbytes = MAX_LFS_FILESIZE;
592    sb->s_blocksize = PAGE_CACHE_SIZE;
593    sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
594    sb->s_magic = DLMFS_MAGIC;
595    sb->s_op = &dlmfs_ops;
596    inode = dlmfs_get_root_inode(sb);
597    if (!inode)
598        return -ENOMEM;
599
600    root = d_alloc_root(inode);
601    if (!root) {
602        iput(inode);
603        return -ENOMEM;
604    }
605    sb->s_root = root;
606    return 0;
607}
608
609static const struct file_operations dlmfs_file_operations = {
610    .open = dlmfs_file_open,
611    .release = dlmfs_file_release,
612    .poll = dlmfs_file_poll,
613    .read = dlmfs_file_read,
614    .write = dlmfs_file_write,
615};
616
617static const struct inode_operations dlmfs_dir_inode_operations = {
618    .create = dlmfs_create,
619    .lookup = simple_lookup,
620    .unlink = dlmfs_unlink,
621};
622
623/* this way we can restrict mkdir to only the toplevel of the fs. */
624static const struct inode_operations dlmfs_root_inode_operations = {
625    .lookup = simple_lookup,
626    .mkdir = dlmfs_mkdir,
627    .rmdir = simple_rmdir,
628};
629
630static const struct super_operations dlmfs_ops = {
631    .statfs = simple_statfs,
632    .alloc_inode = dlmfs_alloc_inode,
633    .destroy_inode = dlmfs_destroy_inode,
634    .clear_inode = dlmfs_clear_inode,
635    .drop_inode = generic_delete_inode,
636};
637
638static const struct inode_operations dlmfs_file_inode_operations = {
639    .getattr = simple_getattr,
640    .setattr = dlmfs_file_setattr,
641};
642
643static int dlmfs_get_sb(struct file_system_type *fs_type,
644    int flags, const char *dev_name, void *data, struct vfsmount *mnt)
645{
646    return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
647}
648
649static struct file_system_type dlmfs_fs_type = {
650    .owner = THIS_MODULE,
651    .name = "ocfs2_dlmfs",
652    .get_sb = dlmfs_get_sb,
653    .kill_sb = kill_litter_super,
654};
655
656static int __init init_dlmfs_fs(void)
657{
658    int status;
659    int cleanup_inode = 0, cleanup_worker = 0;
660
661    dlmfs_print_version();
662
663    status = bdi_init(&dlmfs_backing_dev_info);
664    if (status)
665        return status;
666
667    dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
668                sizeof(struct dlmfs_inode_private),
669                0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
670                    SLAB_MEM_SPREAD),
671                dlmfs_init_once);
672    if (!dlmfs_inode_cache) {
673        status = -ENOMEM;
674        goto bail;
675    }
676    cleanup_inode = 1;
677
678    user_dlm_worker = create_singlethread_workqueue("user_dlm");
679    if (!user_dlm_worker) {
680        status = -ENOMEM;
681        goto bail;
682    }
683    cleanup_worker = 1;
684
685    user_dlm_set_locking_protocol();
686    status = register_filesystem(&dlmfs_fs_type);
687bail:
688    if (status) {
689        if (cleanup_inode)
690            kmem_cache_destroy(dlmfs_inode_cache);
691        if (cleanup_worker)
692            destroy_workqueue(user_dlm_worker);
693        bdi_destroy(&dlmfs_backing_dev_info);
694    } else
695        printk("OCFS2 User DLM kernel interface loaded\n");
696    return status;
697}
698
699static void __exit exit_dlmfs_fs(void)
700{
701    unregister_filesystem(&dlmfs_fs_type);
702
703    flush_workqueue(user_dlm_worker);
704    destroy_workqueue(user_dlm_worker);
705
706    kmem_cache_destroy(dlmfs_inode_cache);
707
708    bdi_destroy(&dlmfs_backing_dev_info);
709}
710
711MODULE_AUTHOR("Oracle");
712MODULE_LICENSE("GPL");
713
714module_init(init_dlmfs_fs)
715module_exit(exit_dlmfs_fs)
716

Archive Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9



interactive